a 2021 perspective on edge computing
TRANSCRIPT
ThoughtLeadership
A 2021 perspective on edge computing
White paperScientific Community
02
Executive summary
03A 2021 perspective on edge computing
Edge computing continues to be a major IT trend in 2021
This paper provides a review on the state-of-the-art and todayrsquos edge computing practices specifically focusing on the software tools that enable execution of workloads at the edge Our intention is to introduce the different scenarios for edge computing and analyze the software available
We have provided a review of the existing development frameworks for each type of IoT edge device as well as the open source frameworks and toolsets that enable the development of IoT edge services and applications
Additionally we explore key issues for the future development of edge services such as the operation of services at the edge and the status of edge standardization efforts
Finally we provide our view on the complexity that edge computing brings to the IT landscape and outline our perspective on how edge computing may evolve
Introduction
The emergence of edge computing is linked to the need to process data close to data generation sources in IoT devices According to the latest estimates we could see more than 7544 billion IoT devices by 20251 mdash a 5X increase since 2015 In this context the challenge is not simply managing IoT devices but more importantly how to cope with the volume of data and content that connected devices will produce
In todayrsquos prevailing IoT architecture sensor data is transmitted over a wide area network to be centralized processed and analyzed mdash which creates an additional supply of enriched data These data and analytical models are intended to trigger actions either on the thing itself in upstream business systems or in other platforms that can use the data outside of the original context
Unfortunately this IoT approach is unsustainable in the long term due to the number of device connections volume of data latency across different locations and networks and the asynchronous nature of many connections between data flow and analytical cloud services In addition todayrsquos heterogeneous networks are unable to manage the massive growth anticipated in the number of endpoints and the data volume Finally smart ways to distribute the data are not yet available
These challenges have created a pressing need to move information and analytical models closer to the source of data in order to provide compute capability inside an environment where connectivity and response times can be controlled To address these needs a new class of edge computing and connectivity has emerged IDC predicts that
At present IoT devices are not only proliferating but simultaneously increasing in sophistication The compute demands of artificial intelligence (AI) processes demand that future IoT environments are composed of more than simple sensors with 8-bit microprocessors
Gradually they will be populated by devices capable of porting diverse sensors and actuators combined with heterogeneous (general purpose and GPU) microprocessors In fact IDC expects this trend to be the main source of growth for the microprocessor industry with edge devices representing 405 of the market by 20233
The proliferation of these AI-powered rich IoT edge devices (the so-called ldquoempowered edgerdquo 4) will enable the presence of an increasingly growing computing continuum mdash ranging from the cloud to numerous devices at the edge referred to as ldquoautonomous thingsrdquo by business analysts like Gartner34
This paper will provide deeper insight into state-of-the-art of edge and edge computing with a specific emphasis on the software tools which enable execution of workloads at the edge Our goal is to identify the technologies that enable edge implementations understand the market adoption status and highlight the technologies and vendors that mdash in our opinion mdash will lead the way
With this in mind we analyzed existing edge computing software frameworks from two perspectives (1) a device perspective providing a study of existent development frameworks by the type of IoT edge device and (2) open source frameworks and toolsets which can facilitate the development of IoT edge solutions Finally we will highlight the fundamental challenges of managing the growing complexity that edge computing brings to the IT landscape
Before we dive deeper letrsquos first define a few terms Edge computing according to the Open Glossary5 refers to the delivery of computing capabilities to the logical extremes of a network in order to improve the performance operating cost and reliability of applications and services By shortening the distance between devices and the cloud resources that serve them and reducing network hops edge computing mitigates the latency and bandwidth constraints of todayrsquos Internet ushering in new classes of applications Technology vendors (data center OEMs and software manufacturers) take the edge opportunity in order to push to the market IoT gateways edge servers edge computing platforms and solutions Every endpoint and its associated edge infrastructure become a de facto extension of the cloud As a consequence overall management complexity increases substantially By the same token the cloud is also transformed because the stream of data is included in a more complex workflow to orchestrate adding real-time constraints to decision making to loop back to the edge For the purposes of clarity when we refer to ldquothe edgerdquo in this document we define it as a heterogeneous set of edge components such as edge devices edge gateways and edge servers interlinked by edge connectors
ldquoBy 2023 over 50 of new enterprise IT infrastructure deployed will be at the Edge rather than corporate datacenters up from less than 10 today by 2024 the number of apps at the Edge will increase 800rdquo 2
04
Figure 1 Classification of edge servers
Edge devices are able to collect and process data from physical things in the field Edge devices incorporate transducers to connect with the physical world (sensors andor actuators) some processing capability (memory microcontroller AI accelerators) and some sort of communications An edge device has also a uniquely identifiable address for the given network
More and more edge devices are gaining processing capabilities by incorporating not only microcontrollers but full-fledged microprocessors The execution of AI workloads is the main driver for these developments Existing developments are expected to be surpassed soon thanks to existing efforts in defining neuromorphic edge processors specifically designed for edge execution of neural networks and other AI workloads
The increasing computing demand driven by the need to provide more intelligent edge features is prompting the emergence of innovative sets of compute boards designed to be embedded into real life objects
Edge servers according to Gartner ldquocollect data deliver content and perform analytics and decision making close to data producers (eg sensors and cameras) and data consumers (eg people and actuators) They can range from standard servers configured for edge processing (deployed in a data centercloset) to micro data centers that have self-contained power and cooling (deployed anywhere)rdquo 6 Figure 1 represents the diversity of current edge servers and highlights their different capabilities
Edge gateways (or IoT gateways) are intermediate devices between edge devices and the applications that create value from their data and access Edge gateways allow the collection and secure transport of data from devices remote users and applications Edge servers with adequate software can act as edge gateways and certain edge gateway devices include enough capacity to host edge services
Edge connectors represent the different connectivity options (usually other than TCPIP) between edge devices and gateways in a wide range of specifications like bandwidth power consumption and range
Hybrid cloud
Connect enterprisesolutions
Run cloudapplications locally
Supportfor virtualization and standart OS
Supportfor standardframework
Complete edge data center
Collect andtrain models
Complex modelexecution
Complexframework
support
Virtualizationsupport
Limited communication
support
Support for basic OS or RTOS
Hardware specifictuning and deployment
Simple framework support
Direct IO and constrained
compute
Single purposefunctionality
Single chip computers (microcontrollers)
Cost eective edge server Devices based on multi processors with limited processing (processing boards)
Edge server Edge devices with moderate compute capabilities
BullSequana SA20GBullSequana EdgeBullSequana Edge nano
Edge high-end server Edge devices based on multi processors high-end compute in CPUs and GPUs
Edge datacenter or micro datacenter very high end compute capabilities at edge
BullSequana X
05A 2021 perspective on edge computing
Edge data handling and security
bull Distribution only In this scenario edge devices provideconsumers with a closer access point for large amountsof content Data is static or changes infrequently froma central control point Data handling is focused primarily on fastdistribution over a wide geography
bull Simple processing Provides connectivity to high velocity datain order to aggregate data qualify value monitor certainpatterns to reduce bandwidth and to bring the right datato cloud for additional processing This category of edgesolutions is often involved in pre-processing data
bull Complex processing Edge solutions in this categorypre-process data before sending it to cloud It also bringsadditional cloud-based business logic to the edge which can
take the form of applications for controlling assets or analytical models mdash including complex business rules Some examples include handling video data and operating in disconnected environments or industrial environments where a near real-time response is required based on the data
bull Multi-application processing This category not only supportscomplex processing but also supports the deployment of entiredata platforms including multiple applicationsProcessing and consumption of data is generally handled fromthe edge These are high-end edge data centers that providelocalized completely independent solutions that can be partof a hybrid cloud setup Edge devices in this category are morecomplex in nature generally multi-node hardware witha virtualization layer
Every class of edge device has different processing capabilities in terms of compute and storage This ranges from running embedded programs for device controls performing business rule execution to executing models that support complete business solution platforms We classify data handling into the following broad categories
One of the consequences of edge adoption is the need for new security paradigms and controls in order to protect data and applications at the edge Existing security mechanisms do not work well with distributed architectures computing spread among multiple locations and when most data is created processed and consumed at the edge without entering the central data center perimeter
Edge forces security to be more efficient than ever and edge security controls must incorporate intelligence Classic architectures typically benefit from ldquodefense-in-depthrdquo approaches where multi-layered security controls protect the data hidden at the back-end Such architectures can withstand some controls being defeated or having mispatchedmisconfigured systems while still providing a high degree of confidence and resiliency because other security layers provide assurance
With the advent of edge businesses can no longer afford such an approach There is a need for more dynamic security controls that are able to adapt to heterogeneous environments without centralized monitoring and administration
One clue to how the cybersecurity market is rapidly adapting to edge is the way analysts talk about the topic Gartner recently coined the term ldquoSASErdquo mdash which stands for Secure Access Service Edge SASE describes a new network security model that combines multiple security controls such as Zero Trust Network Access (ZTNA) cloud access security broker (CASB) firewall as a service (FWaaS) data loss protection (DLP) and more to provide assurance to the ongoing transformation of networking and security in the cloud
06
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
02
Executive summary
03A 2021 perspective on edge computing
Edge computing continues to be a major IT trend in 2021
This paper provides a review on the state-of-the-art and todayrsquos edge computing practices specifically focusing on the software tools that enable execution of workloads at the edge Our intention is to introduce the different scenarios for edge computing and analyze the software available
We have provided a review of the existing development frameworks for each type of IoT edge device as well as the open source frameworks and toolsets that enable the development of IoT edge services and applications
Additionally we explore key issues for the future development of edge services such as the operation of services at the edge and the status of edge standardization efforts
Finally we provide our view on the complexity that edge computing brings to the IT landscape and outline our perspective on how edge computing may evolve
Introduction
The emergence of edge computing is linked to the need to process data close to data generation sources in IoT devices According to the latest estimates we could see more than 7544 billion IoT devices by 20251 mdash a 5X increase since 2015 In this context the challenge is not simply managing IoT devices but more importantly how to cope with the volume of data and content that connected devices will produce
In todayrsquos prevailing IoT architecture sensor data is transmitted over a wide area network to be centralized processed and analyzed mdash which creates an additional supply of enriched data These data and analytical models are intended to trigger actions either on the thing itself in upstream business systems or in other platforms that can use the data outside of the original context
Unfortunately this IoT approach is unsustainable in the long term due to the number of device connections volume of data latency across different locations and networks and the asynchronous nature of many connections between data flow and analytical cloud services In addition todayrsquos heterogeneous networks are unable to manage the massive growth anticipated in the number of endpoints and the data volume Finally smart ways to distribute the data are not yet available
These challenges have created a pressing need to move information and analytical models closer to the source of data in order to provide compute capability inside an environment where connectivity and response times can be controlled To address these needs a new class of edge computing and connectivity has emerged IDC predicts that
At present IoT devices are not only proliferating but simultaneously increasing in sophistication The compute demands of artificial intelligence (AI) processes demand that future IoT environments are composed of more than simple sensors with 8-bit microprocessors
Gradually they will be populated by devices capable of porting diverse sensors and actuators combined with heterogeneous (general purpose and GPU) microprocessors In fact IDC expects this trend to be the main source of growth for the microprocessor industry with edge devices representing 405 of the market by 20233
The proliferation of these AI-powered rich IoT edge devices (the so-called ldquoempowered edgerdquo 4) will enable the presence of an increasingly growing computing continuum mdash ranging from the cloud to numerous devices at the edge referred to as ldquoautonomous thingsrdquo by business analysts like Gartner34
This paper will provide deeper insight into state-of-the-art of edge and edge computing with a specific emphasis on the software tools which enable execution of workloads at the edge Our goal is to identify the technologies that enable edge implementations understand the market adoption status and highlight the technologies and vendors that mdash in our opinion mdash will lead the way
With this in mind we analyzed existing edge computing software frameworks from two perspectives (1) a device perspective providing a study of existent development frameworks by the type of IoT edge device and (2) open source frameworks and toolsets which can facilitate the development of IoT edge solutions Finally we will highlight the fundamental challenges of managing the growing complexity that edge computing brings to the IT landscape
Before we dive deeper letrsquos first define a few terms Edge computing according to the Open Glossary5 refers to the delivery of computing capabilities to the logical extremes of a network in order to improve the performance operating cost and reliability of applications and services By shortening the distance between devices and the cloud resources that serve them and reducing network hops edge computing mitigates the latency and bandwidth constraints of todayrsquos Internet ushering in new classes of applications Technology vendors (data center OEMs and software manufacturers) take the edge opportunity in order to push to the market IoT gateways edge servers edge computing platforms and solutions Every endpoint and its associated edge infrastructure become a de facto extension of the cloud As a consequence overall management complexity increases substantially By the same token the cloud is also transformed because the stream of data is included in a more complex workflow to orchestrate adding real-time constraints to decision making to loop back to the edge For the purposes of clarity when we refer to ldquothe edgerdquo in this document we define it as a heterogeneous set of edge components such as edge devices edge gateways and edge servers interlinked by edge connectors
ldquoBy 2023 over 50 of new enterprise IT infrastructure deployed will be at the Edge rather than corporate datacenters up from less than 10 today by 2024 the number of apps at the Edge will increase 800rdquo 2
04
Figure 1 Classification of edge servers
Edge devices are able to collect and process data from physical things in the field Edge devices incorporate transducers to connect with the physical world (sensors andor actuators) some processing capability (memory microcontroller AI accelerators) and some sort of communications An edge device has also a uniquely identifiable address for the given network
More and more edge devices are gaining processing capabilities by incorporating not only microcontrollers but full-fledged microprocessors The execution of AI workloads is the main driver for these developments Existing developments are expected to be surpassed soon thanks to existing efforts in defining neuromorphic edge processors specifically designed for edge execution of neural networks and other AI workloads
The increasing computing demand driven by the need to provide more intelligent edge features is prompting the emergence of innovative sets of compute boards designed to be embedded into real life objects
Edge servers according to Gartner ldquocollect data deliver content and perform analytics and decision making close to data producers (eg sensors and cameras) and data consumers (eg people and actuators) They can range from standard servers configured for edge processing (deployed in a data centercloset) to micro data centers that have self-contained power and cooling (deployed anywhere)rdquo 6 Figure 1 represents the diversity of current edge servers and highlights their different capabilities
Edge gateways (or IoT gateways) are intermediate devices between edge devices and the applications that create value from their data and access Edge gateways allow the collection and secure transport of data from devices remote users and applications Edge servers with adequate software can act as edge gateways and certain edge gateway devices include enough capacity to host edge services
Edge connectors represent the different connectivity options (usually other than TCPIP) between edge devices and gateways in a wide range of specifications like bandwidth power consumption and range
Hybrid cloud
Connect enterprisesolutions
Run cloudapplications locally
Supportfor virtualization and standart OS
Supportfor standardframework
Complete edge data center
Collect andtrain models
Complex modelexecution
Complexframework
support
Virtualizationsupport
Limited communication
support
Support for basic OS or RTOS
Hardware specifictuning and deployment
Simple framework support
Direct IO and constrained
compute
Single purposefunctionality
Single chip computers (microcontrollers)
Cost eective edge server Devices based on multi processors with limited processing (processing boards)
Edge server Edge devices with moderate compute capabilities
BullSequana SA20GBullSequana EdgeBullSequana Edge nano
Edge high-end server Edge devices based on multi processors high-end compute in CPUs and GPUs
Edge datacenter or micro datacenter very high end compute capabilities at edge
BullSequana X
05A 2021 perspective on edge computing
Edge data handling and security
bull Distribution only In this scenario edge devices provideconsumers with a closer access point for large amountsof content Data is static or changes infrequently froma central control point Data handling is focused primarily on fastdistribution over a wide geography
bull Simple processing Provides connectivity to high velocity datain order to aggregate data qualify value monitor certainpatterns to reduce bandwidth and to bring the right datato cloud for additional processing This category of edgesolutions is often involved in pre-processing data
bull Complex processing Edge solutions in this categorypre-process data before sending it to cloud It also bringsadditional cloud-based business logic to the edge which can
take the form of applications for controlling assets or analytical models mdash including complex business rules Some examples include handling video data and operating in disconnected environments or industrial environments where a near real-time response is required based on the data
bull Multi-application processing This category not only supportscomplex processing but also supports the deployment of entiredata platforms including multiple applicationsProcessing and consumption of data is generally handled fromthe edge These are high-end edge data centers that providelocalized completely independent solutions that can be partof a hybrid cloud setup Edge devices in this category are morecomplex in nature generally multi-node hardware witha virtualization layer
Every class of edge device has different processing capabilities in terms of compute and storage This ranges from running embedded programs for device controls performing business rule execution to executing models that support complete business solution platforms We classify data handling into the following broad categories
One of the consequences of edge adoption is the need for new security paradigms and controls in order to protect data and applications at the edge Existing security mechanisms do not work well with distributed architectures computing spread among multiple locations and when most data is created processed and consumed at the edge without entering the central data center perimeter
Edge forces security to be more efficient than ever and edge security controls must incorporate intelligence Classic architectures typically benefit from ldquodefense-in-depthrdquo approaches where multi-layered security controls protect the data hidden at the back-end Such architectures can withstand some controls being defeated or having mispatchedmisconfigured systems while still providing a high degree of confidence and resiliency because other security layers provide assurance
With the advent of edge businesses can no longer afford such an approach There is a need for more dynamic security controls that are able to adapt to heterogeneous environments without centralized monitoring and administration
One clue to how the cybersecurity market is rapidly adapting to edge is the way analysts talk about the topic Gartner recently coined the term ldquoSASErdquo mdash which stands for Secure Access Service Edge SASE describes a new network security model that combines multiple security controls such as Zero Trust Network Access (ZTNA) cloud access security broker (CASB) firewall as a service (FWaaS) data loss protection (DLP) and more to provide assurance to the ongoing transformation of networking and security in the cloud
06
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Executive summary
03A 2021 perspective on edge computing
Edge computing continues to be a major IT trend in 2021
This paper provides a review on the state-of-the-art and todayrsquos edge computing practices specifically focusing on the software tools that enable execution of workloads at the edge Our intention is to introduce the different scenarios for edge computing and analyze the software available
We have provided a review of the existing development frameworks for each type of IoT edge device as well as the open source frameworks and toolsets that enable the development of IoT edge services and applications
Additionally we explore key issues for the future development of edge services such as the operation of services at the edge and the status of edge standardization efforts
Finally we provide our view on the complexity that edge computing brings to the IT landscape and outline our perspective on how edge computing may evolve
Introduction
The emergence of edge computing is linked to the need to process data close to data generation sources in IoT devices According to the latest estimates we could see more than 7544 billion IoT devices by 20251 mdash a 5X increase since 2015 In this context the challenge is not simply managing IoT devices but more importantly how to cope with the volume of data and content that connected devices will produce
In todayrsquos prevailing IoT architecture sensor data is transmitted over a wide area network to be centralized processed and analyzed mdash which creates an additional supply of enriched data These data and analytical models are intended to trigger actions either on the thing itself in upstream business systems or in other platforms that can use the data outside of the original context
Unfortunately this IoT approach is unsustainable in the long term due to the number of device connections volume of data latency across different locations and networks and the asynchronous nature of many connections between data flow and analytical cloud services In addition todayrsquos heterogeneous networks are unable to manage the massive growth anticipated in the number of endpoints and the data volume Finally smart ways to distribute the data are not yet available
These challenges have created a pressing need to move information and analytical models closer to the source of data in order to provide compute capability inside an environment where connectivity and response times can be controlled To address these needs a new class of edge computing and connectivity has emerged IDC predicts that
At present IoT devices are not only proliferating but simultaneously increasing in sophistication The compute demands of artificial intelligence (AI) processes demand that future IoT environments are composed of more than simple sensors with 8-bit microprocessors
Gradually they will be populated by devices capable of porting diverse sensors and actuators combined with heterogeneous (general purpose and GPU) microprocessors In fact IDC expects this trend to be the main source of growth for the microprocessor industry with edge devices representing 405 of the market by 20233
The proliferation of these AI-powered rich IoT edge devices (the so-called ldquoempowered edgerdquo 4) will enable the presence of an increasingly growing computing continuum mdash ranging from the cloud to numerous devices at the edge referred to as ldquoautonomous thingsrdquo by business analysts like Gartner34
This paper will provide deeper insight into state-of-the-art of edge and edge computing with a specific emphasis on the software tools which enable execution of workloads at the edge Our goal is to identify the technologies that enable edge implementations understand the market adoption status and highlight the technologies and vendors that mdash in our opinion mdash will lead the way
With this in mind we analyzed existing edge computing software frameworks from two perspectives (1) a device perspective providing a study of existent development frameworks by the type of IoT edge device and (2) open source frameworks and toolsets which can facilitate the development of IoT edge solutions Finally we will highlight the fundamental challenges of managing the growing complexity that edge computing brings to the IT landscape
Before we dive deeper letrsquos first define a few terms Edge computing according to the Open Glossary5 refers to the delivery of computing capabilities to the logical extremes of a network in order to improve the performance operating cost and reliability of applications and services By shortening the distance between devices and the cloud resources that serve them and reducing network hops edge computing mitigates the latency and bandwidth constraints of todayrsquos Internet ushering in new classes of applications Technology vendors (data center OEMs and software manufacturers) take the edge opportunity in order to push to the market IoT gateways edge servers edge computing platforms and solutions Every endpoint and its associated edge infrastructure become a de facto extension of the cloud As a consequence overall management complexity increases substantially By the same token the cloud is also transformed because the stream of data is included in a more complex workflow to orchestrate adding real-time constraints to decision making to loop back to the edge For the purposes of clarity when we refer to ldquothe edgerdquo in this document we define it as a heterogeneous set of edge components such as edge devices edge gateways and edge servers interlinked by edge connectors
ldquoBy 2023 over 50 of new enterprise IT infrastructure deployed will be at the Edge rather than corporate datacenters up from less than 10 today by 2024 the number of apps at the Edge will increase 800rdquo 2
04
Figure 1 Classification of edge servers
Edge devices are able to collect and process data from physical things in the field Edge devices incorporate transducers to connect with the physical world (sensors andor actuators) some processing capability (memory microcontroller AI accelerators) and some sort of communications An edge device has also a uniquely identifiable address for the given network
More and more edge devices are gaining processing capabilities by incorporating not only microcontrollers but full-fledged microprocessors The execution of AI workloads is the main driver for these developments Existing developments are expected to be surpassed soon thanks to existing efforts in defining neuromorphic edge processors specifically designed for edge execution of neural networks and other AI workloads
The increasing computing demand driven by the need to provide more intelligent edge features is prompting the emergence of innovative sets of compute boards designed to be embedded into real life objects
Edge servers according to Gartner ldquocollect data deliver content and perform analytics and decision making close to data producers (eg sensors and cameras) and data consumers (eg people and actuators) They can range from standard servers configured for edge processing (deployed in a data centercloset) to micro data centers that have self-contained power and cooling (deployed anywhere)rdquo 6 Figure 1 represents the diversity of current edge servers and highlights their different capabilities
Edge gateways (or IoT gateways) are intermediate devices between edge devices and the applications that create value from their data and access Edge gateways allow the collection and secure transport of data from devices remote users and applications Edge servers with adequate software can act as edge gateways and certain edge gateway devices include enough capacity to host edge services
Edge connectors represent the different connectivity options (usually other than TCPIP) between edge devices and gateways in a wide range of specifications like bandwidth power consumption and range
Hybrid cloud
Connect enterprisesolutions
Run cloudapplications locally
Supportfor virtualization and standart OS
Supportfor standardframework
Complete edge data center
Collect andtrain models
Complex modelexecution
Complexframework
support
Virtualizationsupport
Limited communication
support
Support for basic OS or RTOS
Hardware specifictuning and deployment
Simple framework support
Direct IO and constrained
compute
Single purposefunctionality
Single chip computers (microcontrollers)
Cost eective edge server Devices based on multi processors with limited processing (processing boards)
Edge server Edge devices with moderate compute capabilities
BullSequana SA20GBullSequana EdgeBullSequana Edge nano
Edge high-end server Edge devices based on multi processors high-end compute in CPUs and GPUs
Edge datacenter or micro datacenter very high end compute capabilities at edge
BullSequana X
05A 2021 perspective on edge computing
Edge data handling and security
bull Distribution only In this scenario edge devices provideconsumers with a closer access point for large amountsof content Data is static or changes infrequently froma central control point Data handling is focused primarily on fastdistribution over a wide geography
bull Simple processing Provides connectivity to high velocity datain order to aggregate data qualify value monitor certainpatterns to reduce bandwidth and to bring the right datato cloud for additional processing This category of edgesolutions is often involved in pre-processing data
bull Complex processing Edge solutions in this categorypre-process data before sending it to cloud It also bringsadditional cloud-based business logic to the edge which can
take the form of applications for controlling assets or analytical models mdash including complex business rules Some examples include handling video data and operating in disconnected environments or industrial environments where a near real-time response is required based on the data
bull Multi-application processing This category not only supportscomplex processing but also supports the deployment of entiredata platforms including multiple applicationsProcessing and consumption of data is generally handled fromthe edge These are high-end edge data centers that providelocalized completely independent solutions that can be partof a hybrid cloud setup Edge devices in this category are morecomplex in nature generally multi-node hardware witha virtualization layer
Every class of edge device has different processing capabilities in terms of compute and storage This ranges from running embedded programs for device controls performing business rule execution to executing models that support complete business solution platforms We classify data handling into the following broad categories
One of the consequences of edge adoption is the need for new security paradigms and controls in order to protect data and applications at the edge Existing security mechanisms do not work well with distributed architectures computing spread among multiple locations and when most data is created processed and consumed at the edge without entering the central data center perimeter
Edge forces security to be more efficient than ever and edge security controls must incorporate intelligence Classic architectures typically benefit from ldquodefense-in-depthrdquo approaches where multi-layered security controls protect the data hidden at the back-end Such architectures can withstand some controls being defeated or having mispatchedmisconfigured systems while still providing a high degree of confidence and resiliency because other security layers provide assurance
With the advent of edge businesses can no longer afford such an approach There is a need for more dynamic security controls that are able to adapt to heterogeneous environments without centralized monitoring and administration
One clue to how the cybersecurity market is rapidly adapting to edge is the way analysts talk about the topic Gartner recently coined the term ldquoSASErdquo mdash which stands for Secure Access Service Edge SASE describes a new network security model that combines multiple security controls such as Zero Trust Network Access (ZTNA) cloud access security broker (CASB) firewall as a service (FWaaS) data loss protection (DLP) and more to provide assurance to the ongoing transformation of networking and security in the cloud
06
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Introduction
The emergence of edge computing is linked to the need to process data close to data generation sources in IoT devices According to the latest estimates we could see more than 7544 billion IoT devices by 20251 mdash a 5X increase since 2015 In this context the challenge is not simply managing IoT devices but more importantly how to cope with the volume of data and content that connected devices will produce
In todayrsquos prevailing IoT architecture sensor data is transmitted over a wide area network to be centralized processed and analyzed mdash which creates an additional supply of enriched data These data and analytical models are intended to trigger actions either on the thing itself in upstream business systems or in other platforms that can use the data outside of the original context
Unfortunately this IoT approach is unsustainable in the long term due to the number of device connections volume of data latency across different locations and networks and the asynchronous nature of many connections between data flow and analytical cloud services In addition todayrsquos heterogeneous networks are unable to manage the massive growth anticipated in the number of endpoints and the data volume Finally smart ways to distribute the data are not yet available
These challenges have created a pressing need to move information and analytical models closer to the source of data in order to provide compute capability inside an environment where connectivity and response times can be controlled To address these needs a new class of edge computing and connectivity has emerged IDC predicts that
At present IoT devices are not only proliferating but simultaneously increasing in sophistication The compute demands of artificial intelligence (AI) processes demand that future IoT environments are composed of more than simple sensors with 8-bit microprocessors
Gradually they will be populated by devices capable of porting diverse sensors and actuators combined with heterogeneous (general purpose and GPU) microprocessors In fact IDC expects this trend to be the main source of growth for the microprocessor industry with edge devices representing 405 of the market by 20233
The proliferation of these AI-powered rich IoT edge devices (the so-called ldquoempowered edgerdquo 4) will enable the presence of an increasingly growing computing continuum mdash ranging from the cloud to numerous devices at the edge referred to as ldquoautonomous thingsrdquo by business analysts like Gartner34
This paper will provide deeper insight into state-of-the-art of edge and edge computing with a specific emphasis on the software tools which enable execution of workloads at the edge Our goal is to identify the technologies that enable edge implementations understand the market adoption status and highlight the technologies and vendors that mdash in our opinion mdash will lead the way
With this in mind we analyzed existing edge computing software frameworks from two perspectives (1) a device perspective providing a study of existent development frameworks by the type of IoT edge device and (2) open source frameworks and toolsets which can facilitate the development of IoT edge solutions Finally we will highlight the fundamental challenges of managing the growing complexity that edge computing brings to the IT landscape
Before we dive deeper letrsquos first define a few terms Edge computing according to the Open Glossary5 refers to the delivery of computing capabilities to the logical extremes of a network in order to improve the performance operating cost and reliability of applications and services By shortening the distance between devices and the cloud resources that serve them and reducing network hops edge computing mitigates the latency and bandwidth constraints of todayrsquos Internet ushering in new classes of applications Technology vendors (data center OEMs and software manufacturers) take the edge opportunity in order to push to the market IoT gateways edge servers edge computing platforms and solutions Every endpoint and its associated edge infrastructure become a de facto extension of the cloud As a consequence overall management complexity increases substantially By the same token the cloud is also transformed because the stream of data is included in a more complex workflow to orchestrate adding real-time constraints to decision making to loop back to the edge For the purposes of clarity when we refer to ldquothe edgerdquo in this document we define it as a heterogeneous set of edge components such as edge devices edge gateways and edge servers interlinked by edge connectors
ldquoBy 2023 over 50 of new enterprise IT infrastructure deployed will be at the Edge rather than corporate datacenters up from less than 10 today by 2024 the number of apps at the Edge will increase 800rdquo 2
04
Figure 1 Classification of edge servers
Edge devices are able to collect and process data from physical things in the field Edge devices incorporate transducers to connect with the physical world (sensors andor actuators) some processing capability (memory microcontroller AI accelerators) and some sort of communications An edge device has also a uniquely identifiable address for the given network
More and more edge devices are gaining processing capabilities by incorporating not only microcontrollers but full-fledged microprocessors The execution of AI workloads is the main driver for these developments Existing developments are expected to be surpassed soon thanks to existing efforts in defining neuromorphic edge processors specifically designed for edge execution of neural networks and other AI workloads
The increasing computing demand driven by the need to provide more intelligent edge features is prompting the emergence of innovative sets of compute boards designed to be embedded into real life objects
Edge servers according to Gartner ldquocollect data deliver content and perform analytics and decision making close to data producers (eg sensors and cameras) and data consumers (eg people and actuators) They can range from standard servers configured for edge processing (deployed in a data centercloset) to micro data centers that have self-contained power and cooling (deployed anywhere)rdquo 6 Figure 1 represents the diversity of current edge servers and highlights their different capabilities
Edge gateways (or IoT gateways) are intermediate devices between edge devices and the applications that create value from their data and access Edge gateways allow the collection and secure transport of data from devices remote users and applications Edge servers with adequate software can act as edge gateways and certain edge gateway devices include enough capacity to host edge services
Edge connectors represent the different connectivity options (usually other than TCPIP) between edge devices and gateways in a wide range of specifications like bandwidth power consumption and range
Hybrid cloud
Connect enterprisesolutions
Run cloudapplications locally
Supportfor virtualization and standart OS
Supportfor standardframework
Complete edge data center
Collect andtrain models
Complex modelexecution
Complexframework
support
Virtualizationsupport
Limited communication
support
Support for basic OS or RTOS
Hardware specifictuning and deployment
Simple framework support
Direct IO and constrained
compute
Single purposefunctionality
Single chip computers (microcontrollers)
Cost eective edge server Devices based on multi processors with limited processing (processing boards)
Edge server Edge devices with moderate compute capabilities
BullSequana SA20GBullSequana EdgeBullSequana Edge nano
Edge high-end server Edge devices based on multi processors high-end compute in CPUs and GPUs
Edge datacenter or micro datacenter very high end compute capabilities at edge
BullSequana X
05A 2021 perspective on edge computing
Edge data handling and security
bull Distribution only In this scenario edge devices provideconsumers with a closer access point for large amountsof content Data is static or changes infrequently froma central control point Data handling is focused primarily on fastdistribution over a wide geography
bull Simple processing Provides connectivity to high velocity datain order to aggregate data qualify value monitor certainpatterns to reduce bandwidth and to bring the right datato cloud for additional processing This category of edgesolutions is often involved in pre-processing data
bull Complex processing Edge solutions in this categorypre-process data before sending it to cloud It also bringsadditional cloud-based business logic to the edge which can
take the form of applications for controlling assets or analytical models mdash including complex business rules Some examples include handling video data and operating in disconnected environments or industrial environments where a near real-time response is required based on the data
bull Multi-application processing This category not only supportscomplex processing but also supports the deployment of entiredata platforms including multiple applicationsProcessing and consumption of data is generally handled fromthe edge These are high-end edge data centers that providelocalized completely independent solutions that can be partof a hybrid cloud setup Edge devices in this category are morecomplex in nature generally multi-node hardware witha virtualization layer
Every class of edge device has different processing capabilities in terms of compute and storage This ranges from running embedded programs for device controls performing business rule execution to executing models that support complete business solution platforms We classify data handling into the following broad categories
One of the consequences of edge adoption is the need for new security paradigms and controls in order to protect data and applications at the edge Existing security mechanisms do not work well with distributed architectures computing spread among multiple locations and when most data is created processed and consumed at the edge without entering the central data center perimeter
Edge forces security to be more efficient than ever and edge security controls must incorporate intelligence Classic architectures typically benefit from ldquodefense-in-depthrdquo approaches where multi-layered security controls protect the data hidden at the back-end Such architectures can withstand some controls being defeated or having mispatchedmisconfigured systems while still providing a high degree of confidence and resiliency because other security layers provide assurance
With the advent of edge businesses can no longer afford such an approach There is a need for more dynamic security controls that are able to adapt to heterogeneous environments without centralized monitoring and administration
One clue to how the cybersecurity market is rapidly adapting to edge is the way analysts talk about the topic Gartner recently coined the term ldquoSASErdquo mdash which stands for Secure Access Service Edge SASE describes a new network security model that combines multiple security controls such as Zero Trust Network Access (ZTNA) cloud access security broker (CASB) firewall as a service (FWaaS) data loss protection (DLP) and more to provide assurance to the ongoing transformation of networking and security in the cloud
06
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Figure 1 Classification of edge servers
Edge devices are able to collect and process data from physical things in the field Edge devices incorporate transducers to connect with the physical world (sensors andor actuators) some processing capability (memory microcontroller AI accelerators) and some sort of communications An edge device has also a uniquely identifiable address for the given network
More and more edge devices are gaining processing capabilities by incorporating not only microcontrollers but full-fledged microprocessors The execution of AI workloads is the main driver for these developments Existing developments are expected to be surpassed soon thanks to existing efforts in defining neuromorphic edge processors specifically designed for edge execution of neural networks and other AI workloads
The increasing computing demand driven by the need to provide more intelligent edge features is prompting the emergence of innovative sets of compute boards designed to be embedded into real life objects
Edge servers according to Gartner ldquocollect data deliver content and perform analytics and decision making close to data producers (eg sensors and cameras) and data consumers (eg people and actuators) They can range from standard servers configured for edge processing (deployed in a data centercloset) to micro data centers that have self-contained power and cooling (deployed anywhere)rdquo 6 Figure 1 represents the diversity of current edge servers and highlights their different capabilities
Edge gateways (or IoT gateways) are intermediate devices between edge devices and the applications that create value from their data and access Edge gateways allow the collection and secure transport of data from devices remote users and applications Edge servers with adequate software can act as edge gateways and certain edge gateway devices include enough capacity to host edge services
Edge connectors represent the different connectivity options (usually other than TCPIP) between edge devices and gateways in a wide range of specifications like bandwidth power consumption and range
Hybrid cloud
Connect enterprisesolutions
Run cloudapplications locally
Supportfor virtualization and standart OS
Supportfor standardframework
Complete edge data center
Collect andtrain models
Complex modelexecution
Complexframework
support
Virtualizationsupport
Limited communication
support
Support for basic OS or RTOS
Hardware specifictuning and deployment
Simple framework support
Direct IO and constrained
compute
Single purposefunctionality
Single chip computers (microcontrollers)
Cost eective edge server Devices based on multi processors with limited processing (processing boards)
Edge server Edge devices with moderate compute capabilities
BullSequana SA20GBullSequana EdgeBullSequana Edge nano
Edge high-end server Edge devices based on multi processors high-end compute in CPUs and GPUs
Edge datacenter or micro datacenter very high end compute capabilities at edge
BullSequana X
05A 2021 perspective on edge computing
Edge data handling and security
bull Distribution only In this scenario edge devices provideconsumers with a closer access point for large amountsof content Data is static or changes infrequently froma central control point Data handling is focused primarily on fastdistribution over a wide geography
bull Simple processing Provides connectivity to high velocity datain order to aggregate data qualify value monitor certainpatterns to reduce bandwidth and to bring the right datato cloud for additional processing This category of edgesolutions is often involved in pre-processing data
bull Complex processing Edge solutions in this categorypre-process data before sending it to cloud It also bringsadditional cloud-based business logic to the edge which can
take the form of applications for controlling assets or analytical models mdash including complex business rules Some examples include handling video data and operating in disconnected environments or industrial environments where a near real-time response is required based on the data
bull Multi-application processing This category not only supportscomplex processing but also supports the deployment of entiredata platforms including multiple applicationsProcessing and consumption of data is generally handled fromthe edge These are high-end edge data centers that providelocalized completely independent solutions that can be partof a hybrid cloud setup Edge devices in this category are morecomplex in nature generally multi-node hardware witha virtualization layer
Every class of edge device has different processing capabilities in terms of compute and storage This ranges from running embedded programs for device controls performing business rule execution to executing models that support complete business solution platforms We classify data handling into the following broad categories
One of the consequences of edge adoption is the need for new security paradigms and controls in order to protect data and applications at the edge Existing security mechanisms do not work well with distributed architectures computing spread among multiple locations and when most data is created processed and consumed at the edge without entering the central data center perimeter
Edge forces security to be more efficient than ever and edge security controls must incorporate intelligence Classic architectures typically benefit from ldquodefense-in-depthrdquo approaches where multi-layered security controls protect the data hidden at the back-end Such architectures can withstand some controls being defeated or having mispatchedmisconfigured systems while still providing a high degree of confidence and resiliency because other security layers provide assurance
With the advent of edge businesses can no longer afford such an approach There is a need for more dynamic security controls that are able to adapt to heterogeneous environments without centralized monitoring and administration
One clue to how the cybersecurity market is rapidly adapting to edge is the way analysts talk about the topic Gartner recently coined the term ldquoSASErdquo mdash which stands for Secure Access Service Edge SASE describes a new network security model that combines multiple security controls such as Zero Trust Network Access (ZTNA) cloud access security broker (CASB) firewall as a service (FWaaS) data loss protection (DLP) and more to provide assurance to the ongoing transformation of networking and security in the cloud
06
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Edge data handling and security
bull Distribution only In this scenario edge devices provideconsumers with a closer access point for large amountsof content Data is static or changes infrequently froma central control point Data handling is focused primarily on fastdistribution over a wide geography
bull Simple processing Provides connectivity to high velocity datain order to aggregate data qualify value monitor certainpatterns to reduce bandwidth and to bring the right datato cloud for additional processing This category of edgesolutions is often involved in pre-processing data
bull Complex processing Edge solutions in this categorypre-process data before sending it to cloud It also bringsadditional cloud-based business logic to the edge which can
take the form of applications for controlling assets or analytical models mdash including complex business rules Some examples include handling video data and operating in disconnected environments or industrial environments where a near real-time response is required based on the data
bull Multi-application processing This category not only supportscomplex processing but also supports the deployment of entiredata platforms including multiple applicationsProcessing and consumption of data is generally handled fromthe edge These are high-end edge data centers that providelocalized completely independent solutions that can be partof a hybrid cloud setup Edge devices in this category are morecomplex in nature generally multi-node hardware witha virtualization layer
Every class of edge device has different processing capabilities in terms of compute and storage This ranges from running embedded programs for device controls performing business rule execution to executing models that support complete business solution platforms We classify data handling into the following broad categories
One of the consequences of edge adoption is the need for new security paradigms and controls in order to protect data and applications at the edge Existing security mechanisms do not work well with distributed architectures computing spread among multiple locations and when most data is created processed and consumed at the edge without entering the central data center perimeter
Edge forces security to be more efficient than ever and edge security controls must incorporate intelligence Classic architectures typically benefit from ldquodefense-in-depthrdquo approaches where multi-layered security controls protect the data hidden at the back-end Such architectures can withstand some controls being defeated or having mispatchedmisconfigured systems while still providing a high degree of confidence and resiliency because other security layers provide assurance
With the advent of edge businesses can no longer afford such an approach There is a need for more dynamic security controls that are able to adapt to heterogeneous environments without centralized monitoring and administration
One clue to how the cybersecurity market is rapidly adapting to edge is the way analysts talk about the topic Gartner recently coined the term ldquoSASErdquo mdash which stands for Secure Access Service Edge SASE describes a new network security model that combines multiple security controls such as Zero Trust Network Access (ZTNA) cloud access security broker (CASB) firewall as a service (FWaaS) data loss protection (DLP) and more to provide assurance to the ongoing transformation of networking and security in the cloud
06
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
bull Physical or perimeter security The nature of edge computingdeployment opens up new frontiers in which physical securityand virtual security are equally critical Physical threats couldinclude tampering with devices to introduce malware throughphysical access or unintentional actions that damage the deviceand data Both threats must be addressed and are valid for everyclass of edge device Countermeasures range from secure datacenter processes like access controls and audiovideo monitoringto monitoring temperature changes and movement or protectingagainst natural disasters like fires and floods
Edge devices deployed in the ldquowildrdquo outside a controlledenvironment (eg a smart city solution) pose a greater riskCountermeasures must be physically installed on the edge deviceitself to act as a lock For example the Atos BullSequana Edgeincludes physical intrusion sensors which disable the systemif an intrusion is detected
bull Firmware and operating systems The next level of protection(most importantly from software threats) is the devicersquos firmwareoperating system and identity In a world of billions of devicesprotecting the identity of a device that sends data to your cloudservices is imperative False or corrupted data transmittedfrom an impostor device can harm dependent servicesHardware-based root of trust such as that provided by AzureSphere5 ensures that no device can be separated from its identity
Secure boot-signed firmware and disk-level encryption are someadditional measures that can be taken to secure the base of edgecompute solutions Remote update capabilities are the keyto regularly updating operating systems and drivers from OEMsand other vendors
bull Edge applications Security must be considered right from the startof application design and specially formulated for the dataand processes that run on the edge device Critical data likecertificates data configurations and parameters must be protected
Selecting the right secure coding practices for each device class is essential These include ensuring logged data does not contain sensitive information that local data stores are encrypted that credentials and other critical information are not stored in plain text on the device or written to a console configuring users with minimum privileges and securing internal communication between edge components
As new devices and edge solutions emerge it will be an ongoing challenge to stay up-to-date with new measures to secure the end-to-end chain and data handling of edge to cloud communication
bull Edge networks Edge devices are connected today via variousnetwork types like Wi-Fi LORAWAN 4G Sig-fox NB-IOT and soon5G They provide both inter-edge device connectivity as wellas edge-to-cloud connectivity Communication must be securedusing encryption techniques for message transport as wellas for the message itself
Consider a scenario where the cloud platform sends an unsecuredcommand to an edge device to execute some operationTypical ldquoman in the middlerdquo attacks allow an attacker to gainunauthorized control of devices
Depending on the edge device class and capabilities it may bedifficult to deploy additional network inspection tools For examplea small footprint device like Raspberry Pi will not allow advancednetwork monitoring tools due to limited compute availability
Edge devices deployed in critical areas should not be directlyconnected to the Internet but via gateways to avoid directlyexposing the edge devices to external attack For instance edgedevices deployed in a factory should connect to the shop floornetwork and use edge and network gateways to transmit datato from the Internet
Security at the edge can be examined at various levels such as physical internal or network related The security risk areas can be classified as follows even if specific edge devices may demand different types of security measures
07A 2021 perspective on edge computing
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
bull Use a standard language including standard securitysupported by a broad variety of devices This allows easierdevice upgrades replacements and operations (includinglifecycle management)
bull Pay close attention to lifecycle management for devicesoftware Over the long term ensuring compatibility betweendifferent devices and with the cloud may become verychallenging Try to limit the number of combinationsof supported softwarereleases for different devices
bull Take into account the nature of the infrastructure Processingcapabilities may vary so decisions about what can be processedor analyzed where (on the device vs in the cloud) may differfrom device to device In edge computing analysisand intelligence should occur as close to the data sourceas possible to reduce the data transfer volume and processing
latency In turn this reduces transmission costs and increases throughput speed and quality of service (QoS) Real-time control functions must be executed at the edge
bull Create a robust multi-cloud device test environment to supportthe onboarding of new devices Code can be tested locallyand distributed to other devices after ensuring everythingworks correctly
bull Consider the throughput and cloud connectivity when makingdecisions about software distribution lifecycle managementand mdash in the case of development mdash where each type of datashould be processed
bull Employ the Agile development methodology whichis the accepted norm to deliver business value as quickly as possible
Modularity Modular architecture enables you to customize the solutions that need to run on edge Depending on the use case framework modules can be combined to support the modularization required
Manageability By their nature edge devices are generally deployed into inaccessible and widespread areas Managing edge devices remotely with over-the-air (OTA) updates is essential to smoothly upgrade frameworks and deployed solution payloads
Orchestration Edge devices must perform multiple operations in parallel as well as bring a workflow flavor to solutions for ease of integration and adaptation Support for orchestrating message flows and service calls is key
Device optimized
Edge devices do not have the infrastructure scalability of cloud so be aware of the hardware limitations of your devices Edge devices range from single chip MCUs to high-end edge data centers (and every combination in-between) but even in this extreme range physical compute power is a limitation
Deployment flexibility
Containerization is not merely a cloud phenomenon but is well suited to edge Edge frameworks need to support flexible deployment models and the resilient and portable nature of Docker containers makes them ideal for edge scenarios
Security Edge devices are the direct and closest control units for ldquothingsrdquo and high control and access security is required at all levels mdash including access to the physical device access to the OS local data and communications
Communication Edge frameworks should be able to send data from edge to cloud cloud to edge and even edge to edge so you must support communication flows towards ldquothingsrdquo as well as towards the platform Abstraction rather than low-level communication protocols is needed to develop solutions which are focused only on business logic
Support processing
Provide operational capabilities to process data flows at the edge Edge use cases generally require solutions to process (filter compute analyze) data generated at the edge before it is sent to cloud Some frameworks support machine learning models that are optimized for specific hardware
Edge software frameworks
The cloud computing model relies on economies of scale automation and high degrees of resource homogeneity By employing standardized homogeneous hardware platforms cloud services are usually deployed in huge server farms that offer cost effective compute storage and networking resources In contrast edge computing environments tend to be characterized by heterogeneity The expected massive growth in connected IoT devices mdash together with a wide variety of use cases mdash brings diversity at multiple levels Servers at the edge can range from simple microcontrollers embedded into products to micro-data center installations with powerful compute and storage resources Edge devices can be stationary and hard-wired or battery-powered mobile devices The need for battery power has its own strict requirements in terms of optimizing device energy usage and autonomy This heterogeneity is illustrated in the edge device classification shown in Figure 1
Fundamentally edge device software development does not differ from any other type of software development It is all about functions events and triggers However there are several best practices to consider when developing edge device software
Edge frameworks support two main aspects of edge devices First they provide a platform to provision and run workloads near to the source of data with full lifecycle support Second they provide the means to orchestrate such workloads State-of-the-art edge frameworks have the following characteristics and provide features that enable critical device functions
The following sections provide an overview of software frameworks that support the development of solutions based on the classification of edge devices For our analysis we studied the characteristics of software frameworks for microcontroller edge devices as well as software frameworks for mid-size edge servers and edge data center or micro data center devices In addition we will analyze existing open source edge management tools for both edge data centers and mid-size edge computers
08
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
A microcontroller unit (MCU) is a very low capacity device that is generally embedded as a part of a larger ldquothingrdquo Microcontrollers embedded with use-case specific software solutions fall within this category Billions of ldquothingsrdquo mdash from toys to large industrial equipment mdash use microcontrollers today connecting the devices to open up a wide range of business opportunities
Various chip manufacturers and cloud providers are working to incorporate elements like running small computation and analytical models local decision making or support for completely disconnected scenarios A number of available IoT boards provide extremely battery efficient operation along with multiple built-in connectivity options and development solutions Some of these include
bull NXP iMX RT7 with AWS Alexa integration
bull Azure Sphere MCU8
bull Google Edge TPU9
bull Texas Instruments SimpleLink10
bull ARM Cortex Microcontroller Software Interface Standard (CMSIS)11with MBed OS
bull Microchip Advanced Software Framework (ASF)12
bull ATMEL Software Framework13
MCUs are based on customized real-time operating systems (RTOS) like AWS FreeRTOS14 MBed OS15 or TI-RTOS16 and support IoT development using their own software development kits (SDKs) Developers can also use common programming languages like CPython to ease the development
MCUs like Azure Sphere or ARM MBed OS specifically address security aspects that are critical considering the billions of devices connected to the Internet Azure Sphere and ARM MBed OS provide hardware-printed root certificates which secure the device and cloud service connectivity In addition they provide a hardened OS and continuous security patching using cloud service
In other cases like the NXP iMX RT solution the MCU boards are pre-integrated with AWS Alexa enabling them to bring voice-controlled features to any device Voice processing like noise cancelling and connectivity to the AWS IoT core is also supported Google offers Tensorflow lite on microcontrollers like Edge TPU to run analytical workloads on smart devices
Integrated MCU kits SDKs and cloud development support are increasingly making IoT connectivity and edge computation more accessible and easier to integrate
While MCUs can be compared with single chip computers providing simple integration options midsize edge devices are analogous to compute devices supporting multi-core CPUs and operating systems like Linux or Windows
These are essentially miniaturized general-purpose computers supporting displays USB ports cameras and local device storage mdash thus providing support for higher compute solutions
Devices are supported by a wide variety of edge frameworks and programming languages A number of existing open source initiatives are beginning to offer handling data and applications at the edge which are presented in detail in section 32 These open source offerings complement existing solutions from cloud services providers
Public cloud providers like AWS Azure and Google offer their own solutions which provide integration with cloud-based solutions and bring common cloud solution elements to the edge AWS Greengrass supports the execution of AWS Lambda functions on edge and brings AWS ML capabilities to edge devices Similarly Azure provides support for Azure Functions and Azure ML on edge
In addition to specific tools these frameworks enable remote management of solutions that run on top which simplifies operations and updates Azure IoT Edge is open source and provides support for running non-Azure specific modules on top
With solutions like Azure Stack17 and the very latest Azure Arc cloud services on-premises can run their own workloads Custom solutions to run applications or analytical models are still required as some cloud features and services are missing in the on-premises implementation With Azure IoT developments can be simplified in so-called IoT edge modules These can be easily tested and deployed to edge devices in a similar approach IoT edge modules run as Docker containers and can execute in both Windows and Linux devices This is the same model used for AWS Greengrass which uses AWS Lambda functions for development18 in order to easily test and transfer code locally or on the edge device More details about these solutions can be found in a previous Atos Edge Computing Whitepaper19
In addition to edge framework offerings from public cloud providers edge hardware vendors are beginning to offer software frameworks to help customers to exploit compute capacities in their platforms An example of this approach is the NVIDIA EGX Edge Computing Platform20 for the NVIDIA Jetson hardware family The platform is compatible with Kubernetes management framework and AI execution with the goal of offering real-time processing
Atos offers its own edge management framework by means of Codex Smart Edge This solution is presented in detail in section 7 the end of this paper
Analysis of edge software frameworks from a device perspective
Microcontroller frameworks
Mid-size edge computer frameworks
09A 2021 perspective on edge computing
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Core Services include the set of services which communicates and orchestrates the North and South sides These services include persistent data repository for data obtained from South side objects enablement of actuation requests from cloud (North side) to IoT objects (South side) metadata storage of the IoT objects connected to EdgeX Foundry instance and configurability of the instance
Supporting Services including monitoring alert and notification
System Management which enables EdgeX Foundry services lifecycle management (installation upgrade start stop)
In this category we present a number of solutions from cloud hyperscalers which enable the local reproduction of public cloud environments at the Edge These solutions are in fact novel forms of hybrid cloud Hybrid cloud compute is very important in use cases that have a strong requirement for data sovereignty or connectivity limitations where IoT data generated at source needs to be processed efficiently at the edge
Google Anthos21 provides an on-premises application runtime environment which can run workloads near the data source Anthos provides an optimized stack to execute workloads on top of bare metal servers at the edge Anthos relies on GKE on-prem22 and facilitates the creation management and update of Kubernetes clusters in local environments while providing a unified experience for application management across edge internal and public cloud offerings
Microsoft Azure Stack23 offers on-premises Azure services It consists of three main solutions the previously introduced Azure Stack Edge for execution of AI and general purpose applications at the edge Azure Stack HCI for the management of virtual machines in local
infrastructures and Azure Stack Hub that brings the experiences of Azure cloud on-premises permitting disconnected solutions Like Google Anthos it enables customers to develop unified DevOps experiences across diverse edge local and cloud environments
Along these lines AWS has announced the general availability for AWS Outposts24 This solution develops the AWS experience in local infrastructures It offers services for virtual machines and container management databases networking and data analytics
In addition to offerings from major cloud providers much experimentation has demonstrated how open source software management frameworks such as OpenStack can be used at the edge25 typically in 5G network functions virtualization (NFV) scenarios
More complex on-premises architectures based on traditional virtualization of container management are often considered to be ldquoedge data centersrdquo but are excluded from the scope of this paper
This section analyzes some existing open source initiatives that currently revolve around developing solutions for edge computing management It is important to note that this list is not intended to be exhaustive but instead to presents some existing work that we consider of greatest interest
Two different approaches can be found today in these solutions First solutions that enable low-cost compute devices to act as an IoT gateway (EdgeX Foundry Eclipse Kura) and second solutions that bring workload distribution features to the edge (KubeEdge Starlingx)
Our main interest when presenting these solutions is to provide insights about existing developments and baseline solutions that can potentially be used to enrich more complex business-driven use cases
Open source software frameworks for edge devices
1 2
3
At the time of publication EdgeX Foundry26 offers a microservices software platform which enables users to build IoT gateway functionality out of an edge device EdgeX Foundry software components allow you to obtain data from the so-called ldquoSouth siderdquo (IoT objects within the physical realm) with cloud services (ldquoNorth siderdquo) in which data can be stored aggregated analyzed and converted into actionable information It is important to note that the framework also contemplates actuation processes from North side to South side EdgeX Foundryrsquos architecture is structured in five main building blocks
EdgeX Foundry
10
Edge data center frameworks
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Export Services Layer to develop the integration and data distribution to Northbound services
Device Services Layer which provides a set of services to allow interaction with IoT devices Interestingly it offers an extensible mechanism to build EdgeX Foundry Device Service microservice for new devices
4 5
Eclipse Kura has developed an approach similar to EdgeX Foundry27 with the goal of offering a platform for building IoT gateways Eclipse Kura is a Java-based platform which offers diverse device abstractions and constructs basic gateway services using several drivers to encapsulate communication protocols and configuration For data services it employs a policy-driven data publishing system and it offers an API mechanism for communications with cloud services
KubeEdge28 develops an ldquoopen source system for extending native containerized application orchestration capabilities to hosts at Edgerdquo It offers infrastructure services that combine networking application deployment and metadata synchronization among cloud and edge environments It is important to note that overall KubeEdge architecture considers a centralized edge management which is performed from the cloud level The consideration of a cluster of devices at the edge currently requires integration among KubeEdge and standard Kubernetes Its architecture is structured into edge and cloud parts namely EdgeCore and CloudCore
At the edge level EdgeCore offers functionality to manage the lifecycle of applications (pods) at a single node level In addition it allows the configuration of monitoring probes secrets container runtimes and volumes in the device as well as managing interactions among the different edge components and the cloud
CloudCore comprises components for management of the different edge environments and permits the handling of both lifecycle management operations and operation fromto the edge environment In addition it offers device management functionalities making use of Kubernetes Custom Resource Definitions for device description Device instances belong to a model and enable users to get static device data (specifications) and dynamic device data (status)
Starlingx29 describes itself as ldquoa complete cloud infrastructure software stack for the edge used by the most demanding applications in industrial IOT telecom video delivery and other ultra-low latency use casesrdquo Starlingx considers geographically distributed edge sites of diverse sizes governed under a central data center which offers orchestration and synchronization services Its edge virtualization platform includes deployment options for bare metal VM and container-based workloads In addition version 30 includes integration of OpenStack and Kubernetes on dedicated physical services
Starlingx Infrastructure Management includes components for configuration management providing node discovery and nodes configuration abilities host management which defines host interfaces and monitoring service management that takes into account high availability messaging and service monitoring fault management allowing user definition of alarms and events In addition the forthcoming software management tool aims to support node software update and patches handling
In addition Starlingx offers a container platform based on Kubernetes Cluster that includes application management based in Helm as well as integration with private cloud management tools based on OpenStack
Open Stack Starlingx
KubeEdge
Eclipse Kura
11A 2021 perspective on edge computing
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Edge solutions are becoming an essential part of the OTIT landscape delivering critical business outcomes From a service management perspective it is important to receive timely accurate business alerts preventive maintenance alerts and the speed at which new devices become operational
It is not just about keeping the ldquolights onrdquo for edge but the growing demand for data quality Receiving a real-time sensor reading is fantastic but how do you know if itrsquos the right reading What if other sensors show a different interpretation
Keeping edge environments operational and ensuring data is processed appropriately creates some specific challenges
bull With hundreds or thousands of edge devices gateways and servers plus a potentially heterogeneous technical landscape lifecyclemanagement can quickly become complex
bull Some devices can be difficult to reach or in hostile environments such as roadway infrastructure nuclear power plants or offshore windfarms
bull Ownership integrity and quality of data must be ensured in environments that are typically exposed In addition potentially huge datavolumes will require effective data retention and storage
Figure 2 Service ecosystem for managed IoT services
Operating the edge
Life cycle management
Edge hardware
Edge runfirme and applications
ManagedIoT edgeservices
Data science and modeling
at the edge
Edge connectivity
Edge security
Full stack monitoring
OTA updates
Choice of servicepackages and levels
Ecosystem of partners
247 Support
Performance scalabilityavailabilitysecurity and compliance
Solutions adaptable to business outcomes
Full control of devicesand gateways
Global presence and technical expertise
12
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
All aspects of edge mdash including hardware applications data and connectivity mdash need to be managed Deep and diverse technical expertise is required to implement connect and manage secure resilient edge solutions and to address the different challenges during operations In addition a solid ecosystem of trusted partners is essential to provide best-of-breed solutions The ability to provide full control over a complex variety of edge components in a device-agnostic way is challenging but crucial
bull Edge applications may include edge runtime such asa container management layer as well as data analyticsand other business applications or middlewareCustom application operations middleware and interfacemanagement container management third-party applicationsupport operating system patching cloud integrationand application performance monitoring are part of applicationmanagement services
bull Edge data management may be limited to standard datastorage and telemetry data handling or may involve moreadvanced topics such as data analytics video processing or AIcapabilities at the edge Business rules and alert managementcan be part of the required service management
bull Compliance with local regional or global regulations may needto be handled as part of data management servicesand end-to-end security needs to be tackled by consideringoptions such as role-based access management PKI certificateauthentication HSM vulnerability assessment end-to-end dataencryption and secure industrial control systems In most casesboth onsite and remote support is required on a 247 basiswith faster resolution time requirements than most traditionalIT operations Edge management SLAs may focus on complexconcepts such as business continuity and data qualityin addition to or instead of traditional SLA terms suchas availability
There is rarely a one-size-fits-all solution for addressing the above challenges Each facet of the service layer is unique and multiple software solutions are often required to address each function and problem Finding the right tooling skills and processes to solve your most pressing edge computing challenges represents an opportunity for service integrators
In addition to the toolsets previously identified in this paper as edge frameworks a complete service ecosystem for edge management must include additional features for managing the edge Some typical tools required include
Solutions such as such as ARM Pelion30 or Telit31 can help manage edge apps on your hardware of choice when combined with the edge workload management frameworks described in section 3
As outlined in section 2 there are now many MCU hardware devices with certificates embedded for major public cloud platform providers This makes it possible to onboard devices to a platform using a zero-touch provisioning process
Combine traditional infrastructure monitoring (Nagios) with containers (Prometheus) workloads and data (Dynatrace ELK)
When working with cellular networks a connectivity management platform provides important services and management tools that enable you to manage the SIM lifecycle including deployments SIM activation or deactivation changing tariff profiles or switching mobile networks
Device managers
Zero-touch provisioning Monitoring tools
Connectivity management platforms
13A 2021 perspective on edge computing
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Perspectives in edge computing
Standardization for edge
Initiatives related to edge computing concepts term definitions and models
bull NIST Fog Computing Conceptual Model33 NIST Formaldefinition of Edge Computing34
bull IEEE P19341 - Nomenclature and Taxonomy for DistributingComputing Communications and Networking alongthe Things-to-Cloud Continuum35 (former OpenFog works)
bull ISO ISOIEC TR 301642020 Internet of things (IoT)mdash Edge computing36
bull Linux Foundationrsquos LF Edge Open Glossary for EdgeComputing37
bull Industrial Internet Consortium The Edge ComputingAdvantage38
Initiatives presenting reference architectures and more technical approaches
bull ETSI Multi-access Edge Computing (MEC) Frameworkand Reference Architecture39
bull 1934-2018 - IEEE Standard for Adoption of OpenFogReference Architecture for Fog Computing40
The term ldquoedgerdquo is all over the place IoT use cases create massive amounts of data and that data is often required to be processed at some level near its source Edge computing addresses that challenge
So are we witnessing a new architecture paradigm emerging after the ldquomove to the cloudrdquo or are we simply going back to the proven ldquodo the main compute on premisesrdquo model What role was played by the strong movement towards digitalization incentivized by the COVID-19 pandemic
Increasingly sustainability and energy efficiency are critical aspects that warrant analysis in every new technology Studies such as Lean ICT estimate that watching a 10-minute video online consumes as much electricity as a smartphone uses in ten days 41
Optimizing energy consumption in large data centers (such as those providing public cloud services) has been a large area of study and development in the last decade and has achieved important advances like the definition of the PUE (Power Usage Effectiveness) metric42
At the edge highly standardized internet services connect with highly variable hardware and software installations on physical devices that form the endpoints of the IoT On the Internet side of the edge a proven set of protocols can be selected as well as standards and services for computing management and orchestration Beyond the edge a very diverse and specialized ecosystem exists
As long as the scenarios remained static it posed few problems mdash and these could be solved in a single integration effort With the requirements for agile processes and configuration at the edge it looks very different The orchestration of devices networks and services with the exchange of data must at least be manageable and at best highly automated
In order to achieve this interoperability it is essential to complete the (potentially semantical) description of each participant in the edge in machine readable form To counteract this complex fragmentation standards must be developed to address the different dimensions of edge computing
Currently there are several pure standardization efforts underway as well as work on normalizing edge computing definitions and capabilities The scope of these initiatives is quite wide covering everything from a pure networking perspective to workload execution at the edge It is important to note that some of these initiatives make use of the term ldquoFog computingrdquo 32
The following list presents some of the standardization initiatives that are currently ongoing as of the date of publication
14
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
With edge computing installations growing and the new dimension of local connectivity between machines and devices the role of the network will change and have an important impact on performance and structure In addition 5G campus networks will increase shop floor flexibility by eliminating the need to re-cable communication links The feature of software defined networking (SDN) mdash where routes bandwidth and other characteristics are managed dynamically mdash will elevate the edge computing principle in combination with network functions virtualization The exploitation of computing capability inside the network itself will adapt to this flexibility with virtualizing payload processing to follow network configurations and topologies In such a distributed network-edge computing ecosystem the computing demands will reshape and optimize SDN beyond networking characteristics to form a dynamic edge-network computing fabric
With the compute evolution at the edge there is an increasing push to perform machine learning (ML) model training on the edge ML model training is the most resource and time-consuming aspect of the AI execution lifecycle combined with the complexity of cloud to edge synchronization For this purpose approaches such as ldquolearning on the edgerdquo and ldquoon-device learningrdquo are being developed We anticipate significant developments in this direction both from hardware and edge software frameworks
In our opinion the current state of edge computing is an initial step towards an even more decentralized view of computing that fully exploits the emergent computing continuum enabled by the combination of ldquoempowered edgerdquo and ldquoautonomous thingsrdquo
At Atos we use the term ldquoswarm computing43 for this new digital infrastructure encompassing sophisticated IoT endpoints edge and multiple cloud platforms working in continuous cooperation to connect entities in the context of large cooperative and self-organizing applications in the emerging compute continuum
The edge computing paradigm is a driver for technology evolution Several complementary technologies are also emerging such as 5G communications Edge computing is evolving from IoT towards a complement of cloud computing in the form of a distributed cloud and as an intermediate step to the computing continuum and swarm computing44
While some aspects and use cases of edge are quite accepted in the industry the most advanced aspects of edge computing are still on the innovation curve of the hype cycle (see the Gartner Hype Cycle for Edge Computing 2020) The solution provider market must consolidate and elements of standardization still need to mature Against this backdrop we see several tensions that will drive the evolution of the edge technology landscape
bull Hyperscalers dominate the cloud and have the meansand weight to push for their de facto standards to dominatemassive deployments On the other hand an ecosystemof specialized solution providers will end up winning someniches or even specialized market sectors (for example NVIDIAin computer vision) Hardware and software providersand telecom operators will benefit from the shift to the edge
bull While hyperconnectivity driven by 5G is generally consideredan enabler for edge and swarm (by enabling seamlessconnectivity within edge and to the cloud) it can also challengethe need for edge computing in latency or bandwidth-drivenuse cases
bull Due to the very definition of edge itself increased data sourcesand more exposed data will condition edge and swarm adoptionto specific environments How do we ensure that data originatesfrom trusted sources Who owns this data Who manages itand under what rules Proper solutions and conventions mustbe in place to secure data identity integrity and sovereignty
bull Rightsizing How to avoid oversizing the edge and duplicatingcloud and edge resources
bull Pressure for power efficient edge devices The computeper watt must be improved by several levels of magnitude
bull Waste management must consider the devicersquos designmechanisms and parts traceability so they canbe decommissioned and recycled
bull The heterogeneity of devices capable of offering edge servicescreates a need to reconsider consolidated energy efficiencymetrics in data centers due to the existence of edge deviceswhich do not require cooling systems
bull The battery limitations of devices togetherwith the demonstrated impact of network transmissionin energy consumption required of specific edge
Despite standardization and convergence edge brings complexity and ecosystem diversity to the end-to-end value chain which will represent a large opportunity for service integrators Edge will drive the need for service offerings that address the specificities of on-premises and embedded solutions in combination with increased demand for hybrid cloud capabilities
However advances in edge computing bring their own set of challenges
15A 2021 perspective on edge computing
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
The authors would like to thank Gert Prieber for early revisions and comments
Vincent Couteau Chief IP Transaction Counsel vincentcouteauatosnet
Jordi Cuartero (Editor-in-Chief) CTO IoT jordicuarteroatosnet
Ricky El-Qasem Automation Domain Leader Global Head of Application Platforming amp Data Center Transformation rickyel-qasematosnet
Ana Juan Ferrer (Editor-in-Chief) Distinguished Expert Edge Domain anajuanfatosnet
Kedar Joglekar Technical Director (Architect) kedarjoglekaratosnet
Marc Llanes Global Cybersecurity Business Development Director marcllanesatosnet
Martin Pfeil CTO Global Siemens Account amp IDM Germany Distinguished Expert martinpfeilatosnet
Jesus Ranz IoT Device and Connectivity Engineer jesusranzatosnet
Wolfgang Thronicke Architect amp Consultant for RampD Projects Enterprise AI and Mobile Solutions wolfgangthronickeatosnet
Purshottam Purswani CTO-APAC Business amp Platform Solutions purshottampurswaniatosnet
Jose Maria Cavanillas Worldwide IoT Partner Management Director jose-mariacavanillasatosnet
Said Derradji Lead Architect saidderradjiatosnet
Francisco Jose Ruiz Jimenez Technical Architect Manager fjoseruizatosnet
Acknowledgements
About the Authors
16
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
Codex Smart Edge is highly scalable industrial IoT and edge computing platform provided by Atos The platform offers a hardware agnostic software solution for machine-to-machine connectivity and distributed intelligence It has the following features
bull Edge computing capabilities from processing simple logicto complex AIML jobs at the edge for near real-time decisions
bull Distributed mesh which develops an interconnected smartnode mesh to support layered architecture and swarmconnectivity services
bull Support for disconnected mode operation in scenarioswith intermittent or limited connectivity enabling zero data losswith local data buffering and historization
bull Reduced bandwidth needs by using data decimation andmulti-layer aggregation
bull End-to-end orchestration and management thanks to Kubernetesand Mesh auto surveillance
bull ldquoNorthboundrdquo services for connectivity to IoT data platformsand clouds
bull ldquoSouthboundrdquo services to communicate and interactwith IoT field devices supporting a diverse set of protocols
bull Microservices architecture for extensibility and scalabilityincluding Docker-enabled OS extension support
Atos Codex Smart Edge allows organizations to quickly build and deliver interoperability between devices applications and services for many industrial use cases It provides a foundation for industrial IoT and enables interesting development opportunities for the future
Atos Codex Smart Edge includes ready-to-use native assets to quickly compose project applications aligned to business needs Components range from industrial data collection and simple calculations to AI-based edge analytics and scoring data visualization and export to major cloud platforms
Atos Codex Smart Edge provides the flexibility required to securely and efficiently manage industrial operations in real-time from any location
Atos has designed a powerful set of edge servers to address the needs of edge computing from edge datacentercloud to far edge which can be combined with Atos Computer Vision Platform Codex Smart Edge and Edge Data Analytics (predictive analytics use cases)
Atos Codex Smart Edge
Atos edge computing server range
Annex 1Atos edge assets
17A 2021 perspective on edge computing
Atos Computer Vision Platform
Atos Computer Vision Platform is a unique end-to-end computer vision platform providing pre-trained amp customizable AI models powered by BullSequana server range and enriched by Atos computer vision experts through worldwide experts labs
It enables to identify events and behaviours to reduce error rates to guarantee people and asset safety to deliver highest quality to offer frictionless and personalized customer experiences Business and organisations keep up the paste of events and demand by analyzing videos in real time at the edge to drive the best decisions
bull AI expertise throughcomputer vision labs
bull Pre-trained AI models
bull Market leading softwarestack based on Ipsotek
bull GPU-enabled hardware
BullSequana Edge nanoBullSequana Edge
BullSequana SA20G
BullSequana X
Plug amp play analytics in a compact amp ruggedized
server AI inference and training outside the datacenter
AI inference amp traininginside the datacenter
High performance computer vision computing for training
Up to 500cameras
Up to 250cameras
Up to 50camerasUp to 4 cameras
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
This annex includes several Edge Use Cases Atos has developed that illustrate how Edge computing is becoming a reality in the context of different industries
Industry Manufacturing
Customer challenges reliance on manual visual inspections product recalls through defects passing undetected increased production time for rework
Solution Quality inspection 3D vision inspection amp 3D in-line metrology Machine vision inspection systems where defects get classified according to their type and are assigned an accompanying
grade or default Edge enables real time detection and analysis of the camera images to raise the alert of a defect in a production line ML models are executed on the Edge limiting data throughput and limiting latency and response time
Outcomes Reduction in defects during a production assembly reducing time to detect defects through manual inspections increased yield and productivity
Automatic quality inspection
Industry Manufacturing
Customer challenges thousands of alarms generated every minute from machines skilled technician detects key alarms that need attention Increased time between detection of critical alert to reacting to alert thereby increasing the chance of production outage
Solution Edge analytics applications run on the edge close to the production line and can predict in near real time critical events
avoiding interruption to the production cycle and limiting the amounts of stream data uploaded to the cloud
Outcomes Reduced number of defects through early detection of critical events thereby improving production uptime and yield automation of operators work and reduction in the waste of materials used for production line
Industry Transportation Public Sector and Defense or Retail (People analytics using same solution detecting people behavior in retail outlets)
Customer challenge increased traffic through unplanned events unauthorized access to restricted areas Automatic Number Plate Recognition for traffic violations smarter roads for vehicle safety
Solution Smart camera with 4G5G capability with built in analytics trained to detect vehicle behavior transmits traffic violations to an Edge server near the camera Multiple cameras in a stretch of road can send their violations to a single Edge server
Outcome Improved traffic management preventing road accidents capturing traffic violations improved safety for drivers
Industry Multiple industries
Customer challenge Track and measure equipment usage on the ground to ensure regular maintenance Monitoring working conditions lone worker safety
Solution LoRa beacons and Smart Edge software on LoRa gateways Equipment tagged with GPS LoRa or Sigfox transmitting geolocation
data to Smart Edge nodes which convey onto factoryplant Edge Server for local analytics
Outcomes Increase the throughput of the maintenance teamproductivity improve overall operating efficiency overall equipment effectiveness and decrease unscheduled down time
Customer challenge manual inspections and reliance on onsite technicians to identify faults with the processing plant with issues detected too late thereby causing unplanned outages of the facility
Solution Distributed mesh of Smart Edge based on smart nodes to support local micro services for value restitution at field factory and corporate office level Dedicated enclosures for field deployment
Industry Energy amp Utilities
Outcome Mass data collection and aggregation Machine learning and advanced analytics embracing large data variety (including external data sources) Improved visibility of treatment plant with better maintenance schedules to reduce unplanned downtimes
Smart control room Prediction of events and quality
Traffic Analytics
People and Asset Tracking
Process Optimization for Wastewater Treatment plant
Annex 2edge use cases examples
Quality inspection based on video captured by cameras (linear 2D or 3D) imaging (RX or confocal) or electrical sensors
18
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
1 IDC FutureScape Worldwide IT Industry 2020 Predictions httpswwwidccomgetdocjspcontainerId=US45599219 2 IDC FutureScape Worldwide IT Industry 2020 Predictions3 IDC Worldwide Edge and Endpoint AI Processing Forecast 2020ndash2024 httpsemeaidccomgetdocjspcontainerId=US45169219 4 Gartner Identifies the Top 10 Strategic Technology Trends for 2020 httpswwwgartnercomennewsroompress-releases2019-10-21-gartner-identifies-the-top-10-strategic-technology-trends-for-2020 5 Open Glossary of Edge Computing [v210] - httpsgithubcomState-of-the-EdgeglossaryblobmasterPDFsOpenGlossaryofEdgeComputing_2019_v20pdf6 Bob Gill Thomas Bittman Chirag Dekate Gartner Hype Cycle for Edge Computing 2019 httpswwwgartnercomendocuments3956137hype-cycle-for-edge-computing-20197iMX RT1010 Crossover MCU with Armreg Cortexreg-M7 core httpswwwnxpcomproductsprocessors-and-microcontrollersarm-microcontrollersi-mx-rt-crossover-mcusi-mx-rt1010-crossover-mcu-with-arm-cortex-m7-coreiMX-RT1010cid=ad_PRG353609_TAC356834_EETECH_PP06ampgclid=Cj0KCQjwvb75BRD1ARIsAP6LcquHCuNF8_Ku1wwtpTziBFaipLtx5No3phTy6ORaHfmjPWDrRnjGLsEaAjegEALw_wcB 8 Azure Sphere httpsazuremicrosoftcomen-usservicesazure-sphere 9 Edge TPU httpscloudgooglecomedge-tpu 10 Texas Instruments Microcontrollers (MCU) httpswwwticommicrocontrollerssimplelink-mcusoverviewhtml11 Arm Developer CMSIS httpsdeveloperarmcomtools-and-softwareembeddedcmsis 12 Microchip Advanced Software Framework (ASF) httpswwwmicrochipcommplabavr-supportadvanced-software-framework 13 Atmel Software Framework httpsgallerymicrochipcompackages4CE20911-D794-4550-8B94-6C66A93228B8 14 AWS FreeRTOS httpsawsamazoncomfreertos 15 ARM MBed OS httpsosmbedcommbed-os 16 Texas Instruments RTOS httpswwwticomtoolTI-RTOS-MCU 17 httpsazuremicrosoftcom 18 AWS Lambda functions httpawsamazoncom lambda 19 Atos Edge Computing Whitepaper httpsatosnetwp-contentuploads201904Atos_EdgeComputing_WP-webpdf 20 httpswwwnvidiacomen-usdata-centerproductsegx-edge-computing21 Anthos at the Edge httpscloudgooglecomsolutionsanthos-edge 22 GKE on-prem overview httpscloudgooglecomanthosgkedocson-premoverview 23 Azure Stack httpsazuremicrosoftcomen-usoverviewazure-stackoverview 24 AWS Outposts httpsawsamazoncomoutposts 25 OSF Edge Computing httpswwwopenstackorgedge-computing 26 EdgeXFoundry httpswwwedgexfoundryorg27 Eclipse Kura httpswwweclipseorgkura28 KubeEdge httpskubeedgeioen29 StarlingX httpswwwstarlingxio30 ARM Pelion httpswwwpelioncom 31 Telit httpswwwtelitcom32 Typically these make use of the differentiation among Edge and Fog terms made in httpswwwnebbiolotechwp-contentuploadswhitepaper-fog-vs-edgepdf 33 Fog Computing Conceptual Model httpswwwnistgovpublicationsfog-computing-conceptual-model 34httpswwwnistgovpublicationsformal-definition-edge-computing-emphasis-mobile-cloud-and-iot-composition C Mahmoudi F Mourlin and A Battou ldquoFormal definition of edge computing An emphasis on mobile cloud and IoT compositionrdquo 2018 Third International Conference on Fog and Mobile Edge Computing (FMEC) Barcelona 2018 pp 34-42 doi 101109FMEC20188364042 35 P19341 - Nomenclature and Taxonomy for Distributing Computing Communications and Networking along the Things-to-Cloud Continuum httpsstandardsieeeorgproject1934_1html 36 ISOIEC TR 301642020 Internet of things (IoT) mdash Edge computing httpswwwisoorgstandard53284html 37 Linux Foundationrsquos LF Edge httpswwwlfedgeorgprojects-old__trashedopenglossary 38 IIC The Edge Computing Advantage httpswwwiiconsortiumorgpdfIIC_Edge_Computing_Advantages_White_Paper_2019-10-24pdf 39 ETSI MEC Multi-access Edge Computing (MEC) Framework and Reference Architecture httpswwwetsiorgdeliveretsi_gsMEC001_099003020101_60gs_MEC003v020101ppdf 40 1934-2018 - IEEE Standard for Adoption of OpenFog Reference Architecture for Fog Computing httpsstandardsieeeorgstandard1934-2018html 41 The Shift project LEAN ICT TOWARDS DIGITAL SOBRIETY httpstheshiftprojectorgwp-contentuploads201903Lean-ICT-Report_The-Shift-Project_2019pdf 42 About PUE and DCiE httpsdcimsupportecostruxureitcomhcen-usarticles360039292493-About-PUE-and-DCiE43 Swarm Computing Concepts technologies and architecture httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf 44 httpsatosnetwp-contentuploads201812atos-swarm-computing-white-paperpdf
References
19A 2021 perspective on edge computing
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos
White Paper
CT-210610-RY-J3389-ATOS-WP-PERSPECTIVE
Find out more about us atosnet atosnetcareers
Letrsquos start a discussion together
About AtosAtos is a global leader in digital transformation with 105000 employees and annual revenue of over euro 11 billion
European number one in cybersecurity cloud and high performance computing the Group provides tailored end-to-end solutions for all industries in 71 countries A pioneer in decarbonization services and products Atos is committed to a secure and decarbonized digital for its clients Atos operates under the brands Atos and Atos|Syntel Atos is a SE (Societas Europaea) listed on the CAC40 Paris stock index
The purpose of Atos is to help design the future of the information space Its expertise and services support the development of knowledge education and research in a multicultural approach and contribute to the development of scientific and technological excellence Across the world the Group enables its customers and employees and members of societies at large to live work and develop sustainably in a safe and secure information space
For more information atosnetscientific-community
Atos the Atos logo Atos | Syntel and Unify are registered trademarks of the Atos group August 2021 copy Copyright 2021 Atos SE Confidential information owned by Atos to be used by the recipient only This document or any part of it may not be reproduced copied circulated andor distributed nor quoted without prior written approval from Atos